The neural mechanisms of vulnerability to suicide in old age remain largely unknown, in part due to a paucity of empirical research. Although post-mortem and imaging studies have yielded a number of candidate neural markers, they are not connected to psychological and cognitive pathways to suicide. Our lab attempts to bridge this gap by investigating decision processes that lead to suicidal behavior. Suicide, especially in old age, is often viewed as a strategic, rational decision. In contrast, we argue that the suicidal criis often escalates unexpectedly, and catastrophic choices result from a failure to appropriately integrate moment-to- moment experiences with pre-existing knowledge and values. Empirically, such faulty integration is seen in people with a history of attempted suicide, who tend to make poor choices based on moment-to-moment feedback in reward/punishment-based learning experiments. The applicant's K23-supported preliminary study combining fMRI and computational modeling found that the suicidal persons' failure to integrate accumulating feedback is paralleled by disrupted paralimbic expected reward signals. The proposed study goes a step further by examining dynamic decision processes in suicidal people at multiple levels: reward signals measured through BOLD activation, corticostriatothalamic network integrity measured by functional connectivity, behavior during reward learning, dimensions of cognitive performance implicated in decision- making, and psychological traits relevant to real-world decisions. Our first aim is to investigate corticolimbic reward system alterations underlying biased decision-making in suicidal behavior. We focus on the neglect of vital information in decisions, test its association with attempted suicide and examine its neural underpinnings. Our second aim explores the well-established link between poor executive control and suicidal behavior. We will test a putative pathway from disrupted neocortical top-down control over decision-making to failure of behavioral adaptation to suicidal ideation. We propose a cross-sectional case-control study with three groups of depressed participants (30 suicide attempters, 30 suicide ideators, 30 non-suicidal depressed elderly) and 30 non-psychiatric controls, recruited from an ongoing study of late-life suicidal behavior (PI: Szanto). Participants will complete a novel probabilistic reversal learning fMRI task, which experimentally manipulates multiple theoretically predicted signals (expected reward, reward probability, reward prediction error, volatility). Analyses of neural data will use a reinforcement learning model. The sampling strategy and detailed assessments are designed to minimize and control for confounding factors. The investigative team includes experts in functional (Siegle) and specifically geriatric (Aizeinstein) imaging, statistical analysis of neural data (Kass), decision neuroscience (Clark), and cognitive control (MacDonald). This work leverages powerful theoretical frameworks of formal learning theory and decision neuroscience. It will lay the groundwork for a dimensional classification of suicidal behavior and for discovery of intervention targets.